[dm-devel] [PATCH 0/9] multipath-tools: avoid I/O during uevent processing

[Martin Wilck]

Hi Christophe, hi Ben,

the bulk of this series attempts to avoid unnecessary I/O for "multipath -u"
calls from udev rules. I came up with it in a recent attempt to debug
avoid "multipath -u" hangups during boot. I found that the detect_alua()
call in the get_refwwid->pathinfo->scsi_ioctl_pathinfo call sequence would
cause an sg ioctl to be sent for every "multipath -u" invocation. That's
obvious nonsense, we don't need to detect ALUA capabilities to find out
if a device is a potential multipath member.

So, rather than setting pp->tpgs in pathinfo, I resorted to a "lazy"
approach where ALUA capabilities are probed when they are needed, no
sooner. I also cleaned up the ALUA code a bit (still some more to go),
and made it attempt to read INQUIRY data from sysfs (this is fine for
the ALUA prioritizer - it only needs to bypass sysfs for the RTPG calls).

Martin Wilck (9):
  libmultipath: add sysfs_get_inquiry()
  libmultipath: alua: make API more consistent
  libmultipath: alua: try to retrieve inquiry data from sysfs
  libmultipath: constify sysfs_get_timeout()
  libmultipath: detect_alua(): use system timeout
  libmultipath: lazy tpgs probing
  libmultipath: don't link alua_rtpg.o twice
  libmultipath: check_rdac(): pre-check in hwtable
  libmultipath: hwtable: add Lenovo DE series

 libmultipath/discovery.c              | 46 +++++++++++++++++++--------
 libmultipath/discovery.h              | 10 +++---
 libmultipath/hwtable.c                | 20 ++++++++++++
 libmultipath/prioritizers/Makefile    |  3 --
 libmultipath/prioritizers/alua.c      |  4 +--
 libmultipath/prioritizers/alua_rtpg.c | 45 ++++++++++++++++++++------
 libmultipath/prioritizers/alua_rtpg.h |  7 ++--
 libmultipath/propsel.c                | 31 ++++++++++++------
 multipathd/main.c                     |  7 ++--
 9 files changed, 126 insertions(+), 47 deletions(-)

-- 
2.21.0